What properties of a multivariable function f(x,y) makes a linearization over or underestimate the true value of a point?

20 Views Asked by At

I understand how a linearization of a multivariable function creates a tangent plane that can be used to estimate points on the actual function and it will either overestimate or underestimate the true value of the function. However, I don't understand what properties of the function allows this to consistently happen. I know in the 2d plane concavity of the function determines this with the second derivative test, can a similar idea be used here?